Univa first announced the "200 bug fixes and enhancements" statement
at SC11 last year. And most likely Univa is comparing UGE against SGE
6.2u5 - but we all know that http://gridengine.sunsource.net is not
even online anymore, so basically anything is better than a dead
project.

Since Oracle passes the Open Grid Scheduler project maintainership of
the open source code base, we added our own improvements. We took what
Oracle gave us and develop our own features & fixes.

Software can never be bug-free, and it is the maintainer's job to
decide how developers' time are allocated. We need a balance between
new features & bug fixes. We give urgent issues the highest priority -
if more than 3 sites hit a bug and send us a report, then we give the
issue a much higher priority. For example, the high profile 4GB memory
accounting bug on Linux was fixed in Open Grid Scheduler/Grid Engine
back in 2010 - way before UGE even existed.

It was also the Open Grid Scheduler project that first implemented
hwloc topology binding, we released our first implementation last
April: http://gridscheduler.sourceforge.net/projects/hwloc/GridEnginehwloc.html

(And if you visit the hwloc homepage, you will see under the "Resource
manager and job schedulers" category, the Open Grid Scheduler project
is listed first - and then Torque, Univa Grid Engine, Son of Grid
Engine added support much later and thus are listed at the bottom.)

And both Univa Grid Engine and Open Grid Scheduler/Grid Engine
released the NVML-based GPU load sensor at around the same time - we
both wanted to release something for SC11 (our implementation was
copied by other Grid Engine projects as well). Also, both benefit from
open source users & developers - eg. William Hay reported the security
bug last Nov and was fixed in all implementations of Grid Engine
before the bug was disclosed to the public; and Univa asked Brooks
Davis the permission to include the qrsh bug fix for BeyondTrust
environments in UGE.

The Open Grid Scheduler project also removed the dependency on NFSv4
for Berkeley DB spooling, and we added other things like getting the
job exit status from epilog via a supported interface, user
notification mails sent from a configurable UID, the ARM Linux port,
gmake upgrade (we released the instruction on the original Sunsource
site back in 2008), and improved support for AIX all in the Grid
Engine 2011.11 release.

I am sure there are features that the Open Grid Scheduler project has
that Univa Grid Engine does not, but I am also sure that there are
fixes and features that Univa put into UGE and are not available
elsewhere.

And to clarify: I am responding to this email not because I wanted to
bash Univa. Also, I wanted to follow-up with my previous complain - if
you google "Open Grid Scheduler" today, you shouldn't be seeing the
annoying Open Grid Scheduler - Download and Try - Built by Experts" Ad
placed by Univa. We, the Open Grid Scheduler project never copied code
from Univa's source tree, so thanks Univa for taking that Google Ad
down.

Finally, if there is an important missing feature in Open Grid
Scheduler/Grid Engine, please let the list know. We are commerical
open source, so by nature, we give bugs reported by Scalable Grid
Engine customers (
http://www.scalablelogic.com/scalable-grid-engine-support ) the
highest priority, but we also fix issues that are important enough for
open source users - but again, we need to balance our allocated time
between new features & bug fixes.

Rayson



On Thu, May 3, 2012 at 11:13 AM, Chi Chan <[email protected]> wrote:
> I read this article yesterday:
>
> http://www.theregister.co.uk/2012/05/02/univa_grid_engine_8_1/
>
> Univa CEO Gary Tyreman said that Univa did "200 enhancements for the
> 8.0.1 release". What is Univa's definition of an 'enhancement'?
>
> --Chi
> _______________________________________________
> users mailing list
> [email protected]
> https://gridengine.org/mailman/listinfo/users
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users

Reply via email to